Meta is overhauling its artificial intelligence division. The company is creating two new teams: AI Products and AGI Foundations. This move aims to speed up innovation and product development, as reported by Axios. Meta faces intense competition from rivals like OpenAI, Google, and ByteDance.
Chief Product Officer Chris Cox announced the changes in an internal memo on May 27. The restructuring could lead to quicker AI feature integration into Facebook, Instagram, and WhatsApp. For users, this signals Meta’s urgent push to lead in the evolving AI landscape. The company hopes smaller, agile teams will foster quicker development and greater ownership.
New Structure For AI Advancement
The AI Products team, under Connor Hayes, will manage the Meta AI assistant. It will also oversee Meta’s AI Studio for AI bot creation, and AI features across Meta’s apps.
Simultaneously, the AGI Foundations unit will focus on core technologies. Ahmad Al-Dahle and Amir Frenkel will co-lead this unit. Their work includes Llama model development and improving AI reasoning, multimedia, and voice capabilities.
Meta’s Fundamental AI Research (FAIR) group remains mostly separate. However, one multimedia team will join the AGI Foundations unit.
No executives are leaving due to these changes. No jobs are being cut, according to the Axios report. Some leaders have moved from other company parts. Cox explained in the memo that the “new structure aims to give each org more ownership while minimizing (but making explicit) team dependencies.”
This is not Meta’s first AI reorganization. The company made a similar reshuffle in 2023 aimed at speeding efforts.
Addressing Talent And Competitive Headwinds
This reorganization occurs at a critical time for Meta. The company has experienced a significant loss of AI talent, as 11 of the 14 original Llama model creators have left.
Many joined competitors, including Paris-based Mistral AI. Mistral AI, co-founded by former Meta Llama architects, now employs five authors of the original Llama paper published on arXiv. Furthermore, Joëlle Pineau, who led Meta’s FAIR group for nearly eight years, exited in April 2025.
These departures coincide with reported delays in Meta’s largest AI model, Llama 4 Behemoth. The developer community has also given a muted response to its Llama 4 models. Competition is fierce, with advancements from rivals like DeepSeek and Alibaba’s Qwen team.
Navigating A Complex Web Of Challenges
Beyond talent and competition, Meta faces other pressures. The company cut jobs in its Reality Labs division due to operating losses, while still recruiting AI engineers.
Ethically, Meta is tuning Llama 4 models to address what it calls a historical left-leaning bias, attributed to “the types of training data available on the internet.” This initiative occurs as Meta also navigates significant scrutiny over its AI data training practices in Europe.
Meta has been using public user data from Facebook and Instagram for AI training, a move facing opposition despite Meta citing ‘legitimate interest’. Data protection advocates and authorities remain critical, although a court ruled in Meta’s favor regarding an injunction.
Legally, Meta is involved in a major copyright case about its AI training practices. Plaintiffs allege Llama models were trained on pirated books. During a hearing, the judge in the case expressed skepticism about Meta’s “fair use” defense if it significantly harms authors’ rights, questioning how destroying a market for someone’s work without licensing could be fair.
The financial burden of AI development is also evident, with reports suggesting Meta sought external funding for Llama development from other big tech companies.
Despite these issues, Meta is advancing its AI platform ambitions, having previewed its Llama API in April after launching the two smaller versions of its three announced Llama 4 models.
Mark Zuckerberg projected in a recent conversation with Microsoft CEO Satya Nadella that “maybe half the development is going to be done by AI, as opposed to people” in the next year, saying that this share “will just kind of increase from there.” This echoed Nadella, who noted AI was likely responsible for “20%, 30% of the code that is inside of our repos today’ was ‘written by software.”
